Evidence Slopes and Effective Dimension in Singular Linear Models

Kalyaan Rao · 2026

Read Evidence Slopes and Effective Dimension in Singular Linear Models by Kalyaan Rao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-01238.txt.

Description

Kalyaan Rao studies how statisticians weigh evidence for one model over another. Standard shortcuts assume a model's effective size equals its raw count of parameters. That assumption breaks for singular models, which are overbuilt or rank-deficient, where a quantity from singular learning theory gives the true effective size instead.

Rao works with linear models where the exact evidence can be computed in closed form, so errors can be measured rather than guessed. He shows the standard shortcut's error grows in a clean pattern set by the gap between raw size and effective size. A correction that uses the effective size restores the right evidence slope and ignores mere re-labelings of the same model. Tests back the theory. The work gives a concrete picture of where the common shortcut fails and offers evidence slopes as a practical way to estimate effective size.

Catalog record

CREATORKalyaan Rao
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-01238

Browse the full library · About the Tower of Babel